The NDIS

The NDIS provides support to people with permanent and significant disability who enter the scheme before age 65. People with disability usually need their health care provider to prepare evidence stating that their disability is permeant and significant. After gathering the information, they need to submit the evidence to the NDIA. It is ultimately up to the NDIA to decide if an applicant’s disability is permeant and significant and thus meet the eligibility criteria of entering the NDIS. Unlike the previous system, which assumes a one-size-fits-all approach and funds services providers directly, the NDIS considers each individual’s personal goals and preferences, then tailors and provides funds for each individual to spend on the services they need. Three types of support budgets can be funded in one’s NDIS plan: the core supports budget, the capacity building supports budget, and the capital supports budget. The core supports budget, in particular, funds assistance such as self-care activities, household tasks, therapeutic supports not covered by the mainstream health system, and transport if one cannot use public transportation. In addition to the government-funded programs available to support carers, an NDIS participant might use funding in their plan to facilitate respite care, which gives carers short breaks from caring responsibilities. All these services are referred to as “formal services” in this paper, services that are professional, regulated, and typically involve monetary transactions.

There have been several stages of the implementation of the NDIS. It was first tested in trial sites in five states/territories in 2013. Prioritisation was given based on age or residential location. The trial ended in July 2016, and since then, the NDIS began to roll out across Australia (except for some early roll-out areas in NSW). ACT was the first state or territory to complete the roll-out, followed by other states and territories, with Western Australia being the last state to reach the full scheme from July 2023. Details regarding the trial and roll-out dates of the NDIS are listed in Appendix S1Table.

There are several steps to receive a support budget from the NDIS. First, one has to apply for the NDIS with the evidence to prove that they meet the eligibility criteria of entry. Then the National Disability Insurance Agency (NDIA) assesses of the minimum eligibility requirements, and will notify the applicant if they are granted access to the NDIS. Planning meetings between an NDIA planner and the participant will take place if access is granted to discuss how much support the participant is likely to need and a plan will be made accordingly. The NDIA then reviews the plan. Once the plan is signed off by the NDIA, participants can arrange their services [13]. The NDIS participants could choose whether their NDIS plan is managed by themselves, by the NDIA or by a plan manager. About 70% of people with disability choose to use a plan manager who helps them with reimbursement and fund monitoring but does not support coordination [14]. About 40% of NDIS participants also have support coordination funded, which helps them coordinate support and services; however, research showed that only about 20% of the participants were able to use at least 80% of their support coordination funding [15]. For those NDIS eligible, the waiting time to get an approved plan could be long, especially during the early stages of the NDIS roll-out. It was estimated that it took more than seven months for a participant to get an approved plan before July 2019 [11], not to mention the time needed to arrange services.

Mechanisms

Fig 2 illustrates the potential mechanisms through which the NDIS roll-out may impact carer outcomes through its impact on formal service utilisation, as the latter is a core funding area of the NDIS. Our study examined the average effect of the NDIS by comparing two key groups – people with profound or severe disability living in areas where the NDIS has been rolled out, and those living in areas where the NDIS has not been rolled out. Given that formal services are a core area funded by NDIS, eligible people living in areas where the NDIS is available could receive more and higher quality formal services and, consequently, the average formal service use in the NDIS available areas may increase (Path A, Fig 2). The literature has competing views on whether increased formal caring hours substitute, complement or are independent of informal caring hours. According to the substitution theory, an increased supply of formal care may decrease informal care [16–20] (Path A1, Fig 2). However, if formal care is provided for tasks that informal carers cannot perform [21] or if it is provided only when the informal care has been exhausted [22], the informal care may remain unchanged even if the formal care increased (Path A2, Fig 2). As highlighted in the NDIS guidelines, the NDIS was not intended to replace informal care; in such circumstances, formal care and informal care may be independent. It is also possible that NDIS increased the informal caring hours as the complementary theory postulates (Path A3, Fig 2). There is not much evidence about the reinforcing effects of formal care. However, more informal care may be needed to permit access to the formal care being provided (e.g., if one needs help to undertake formal care activities). Also, at the early stages of the NDIS roll-out, carers may have needed to spend more time planning and coordinating the approved services for the recipients; therefore, their caring time may have increased.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The potential mechanism through which NDIS may affect caring hours, employment and social participation of primary carers.

https://doi.org/10.1371/journal.pone.0321377.g002

Those ruled ineligible for the NDIS but who may have previously accessed state/territory-based services may now face difficulties in accessing formal services. As a result, the average use of formal services in the area may not change (Path B, Fig 2) or decrease (Path C, Fig 2) even if the NDIS has become available. The caring hours are only expected to change if there is a change in the use of formal services. However, if the formal service use decreases, the caring hours may need to increase to fill this gap.

Given the changes in caring hours, the NDIS may or may not affect labour force and social participation outcomes through the change of caring hours (Path A1 and A3, Fig 2). It is possible that the decreased informal caring hours free up some of the informal carer’s time so that informal carers can work or participate in social events because previous research found that caregiving is associated with higher rates of part-time work and fewer hours of work [23,24]. However, if the NDIS increased informal caring hours, the carers may be less likely to be employed or enjoy the same leisure time simply because they have no free time. There is also a possibility that the NDIS influences employment and social participation outcomes through other channels that are independent of the caring hours. For example, family support and counselling may help the family of people with a disability to manage the demands the disability brings. If mental pressure prevents carers from participating in economic or social activities, then the NDIS may indirectly support carers in this way as well.

This study examines changes occurring within a relatively brief period of 2.5 years maximum ---- a duration considerably shorter compared to the 10-year milestone reached by the NDIS. The impact of the NDIS may be different in the short term compared to the long term. While the impact on caring hours and social participation may show relatively quickly after getting formal care, the effects on employment may take longer to manifest. Also, the impact on employment may depend on the flexibility of available jobs — for example, the number of part-time jobs in the labour market.

Data

Sample

Treatment and control groups.

The NDIS trials and roll-out targeted different population groups. Some states or territories prioritised certain ages, while some were geographical.

Both the treatment and control groups consist of people with profound and severe disability or who need an aid/s who meet the age criteria and their primary carers. The primary carers are aged over 14 years and live in the same private dwelling with their recipient of care. The treatment group are people who lived in the LGAs where the NDIS became available between the two waves of SDAC interviews, i.e., 1 January 2016 and 30 June 2018 (“the treatment period”). In the control group, people lived in the LGAs where the NDIS became available only after the 2018 SDAC interview, i.e., after 1 July 2018 (“the control period”). We exclude people living in the LGAs that received the NDIS before 31 December 2015 as we do not have relevant data before 2015 (i.e., they were always treated). We also excluded people from Western Australia (WA) because WA ran its own trial, which extended into our analysis period, and the NDIS delivery model in the trial differed from the other states and territories.

Details of the timeline are presented in Fig 3, and the details of the trial and roll-out criteria can be found in Appendix S2 Table. Characteristics of the treatment and control groups are presented in the following subsection.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. NDIS Implementation Timeline and Research Design.

https://doi.org/10.1371/journal.pone.0321377.g003

Outcome of interest

As the carer’s outcomes can be changed through the NDIS’ impact on formal services utilisation of the care recipient, this paper focuses on two sets of outcomes: the first set encompasses formal service use by the care recipients and the second set, the outcomes of the primary carers.

For the first set of outcomes at the care recipient level, we use questions about the frequency of assistance received from organised services because of their disability. Health care services are excluded as we cannot distinguish from the survey which ones are covered by the health system (Medicare) versus funded by other means. The survey participants were asked: “On average, how many times per [day/week/month/year]?” if they stated they needed organised care. The frequencies are transformed into ‘per week’ amounts if the original answer was provided in times per day/month/year assuming there are 52 weeks in a year. For example, if the answer is three times per month. Then we assume that it is 36 times (12months *3) per year. This is then converted to 0.69 = 36/52 times/week. We consider the overall numbers of formal services along with the extensive margin (i.e., whether they used any formal services) and the intensive margin (i.e., how many formal services they used when they had used at least some).

In the second set of outcomes at the carer level, our main outcome of interest is the weekly caring hours of primary carers for recipients. We also explore employment and social participation as those will likely change through the change in caring hours. We focus on co-inhabiting primary carers because they tend to care for longer hours, and the NDIS may have the greatest impact on them. Primary carers consist of 35% of all carers in our data, and they were asked: “On average, how many hours do you spend each week providing care to (recipient X)” This variable is converted into a continuous variable from a 6-level categorical variable (<9, 10–19, 20–29, 30–39, 40–69, 60 + hours) using mid-point transformation.

We also analyse the impact on employment and social participation of the primary carers. Our employment indicator is equal to one if the respondent reported they were employed full-time or part-time (and zero if they were unemployed, looking for work or not in the labour force). For social participation, we look at both social participation without the recipient (one if the primary carer had been to certain places or events in the last 12 months without the recipient and zero otherwise) and social participation with or without the recipient.

Further details of how the variables are constructed are listed in Appendix S3Table.

Covariates

We control for several covariates likely to be associated with our outcome of interest but unrelated to NDIS access to increase our estimation precision. The covariates include two sets. The set of carer characteristics includes age, sex, education level and the number of recipients of care. The second set is recipient-level characteristics and includes age, sex, marital status, education level, whether born in Australia, disability severity levels, disability types, the Accessibility Remoteness Index of Australia (ARIA) to reflect the ease or difficulty people face accessing services, the number of adults (aged 15 or more) without disability living in the household, and the number of bedrooms in the household. The last one is regarded as a proxy for the income level of the household, as the income variable has 20% of missing values. Unemployment rates by LGA one year before the start of each wave of survey (September 2014 to June 2015; September 2017 to June 2018) are also controlled for to account for local labour market conditions.

Estimation strategy

To start with, we compare the impact of the NDIS on formal service use of people with profound or severe disability (or needing aids) under 65 living in areas where the NDIS became available between January 2015 to June 2018 and the caring hours, employment and social participation for their primary carers to those with the same characteristics but living where the NDIS was not available until after July 2018 (i.e., after when the 2018 SDAC interview took place). Between the first quarter of 2015 and the second quarter of 2018, the NDIS covers approximately one-fourth of the total population of individuals with severe and profound disabilities in those specific areas. Given that our data does not include information on actual NDIS enrolment, the results represent intent-to-treat (ITT) estimates. ITT estimates are informative for understanding the population-level impact of the policy change. In our case, the results show the change of outcome attributed to the NDIS roll-out (and withdrawal of other services) on all the people with profound or severe disability or needing an aid (aids) who meet the age criteria in the NDIS available areas.

Our identification of the NDIS impact relies on LGA variation in the timing when the NDIS becomes available. Therefore, we need to ensure that the impact of the NDIS on our outcomes is not attributed to other LGA-level factors associated with the availability of the NDIS. We use LGA fixed effects to control for the potential correlation between the NDIS roll-out and time-invariant unobservable LGA characteristics. Simple pre- and post-treatment comparisons could also be influenced by the time trends in the outcome variables and other national policy changes or events; therefore, we then estimate the effect by applying the difference-in-difference (DiD) method using OLS regression. Our specification is as follows:

(1)

For the formal service outcomes, represents the outcome for the overall number of formal services used, the intensive and extensive margin of the formal services of recipient j living in LGA 𝑙 during wave 𝑙. For the carer outcomes, the dependent variable y corresponds to outcomes of carer i of recipient j living in LGA 𝑙; during wave 𝑙. The outcomes include weekly caring hours, employment and social participation. 𝑙𝑙𝑙𝑙_𝑙 equals one if the outcome is measured in 2018 and zero otherwise. T_il equals one if people are in the treatment group and zero if they are in the control group.

The parameter of interest is 𝑙₁, representing the impact of the NDIS on the treatment group relative to the control group after the NDIS became available. This assumes that the differences in the outcomes are attributed to the random-like roll-out of the NDIS participation at the LGA level, not other factors.

We also control for the characteristics of the recipients 𝑙 and the primary carers 𝑙 and the unemployment rate, which are included in vector Z_𝑙𝑙𝑙. Time-fixed effects 𝑙_𝑙 control for common nationwide policy changes such as changes in welfare payments, the availability of new medicines and other nationwide trends between the two waves. The LGA fixed effect 𝑙_𝑙; controls for any time-invariant differences across LGAs that may be correlated with our outcomes, such as demographic composition, supplier market, disability coordination facilities, cultural values, or economic conditions which are crucial factors that may impact our outcomes. 𝑙_{𝑙𝑙𝑙𝑙} is the error term. We do not further match our sample based on characteristics such as age and gender within each LGA as suggested in other literature [25] because there are, on average, only 5 observations per LGA.

The previous evaluation reports on NDIS found different effects for different age groups, and positive employment changes were observed in the carers of participants aged 15–24 [10]. Therefore, we also perform a subgroup analysis for recipients aged 15–24 versus those aged 0–14 and 25–64, who are within the NDIS eligible age criteria.

A key assumption in DiD analysis is that in the absence of the treatment, the trends in the outcomes between the treatment and the control groups would have been the same (the parallel trend assumption). In our study, we cannot test this assumption, given that only two waves of data are available. In the following section, we examine the robustness of the results to a different control group—primary carers of recipients who are not eligible for the NDIS (e.g., the disability status of the recipients is not profound or severe, or they are aged 65 years and older) and they live in same LGAs where the NDIS has not been rolled out.

Given that repeated cross-sectional data is used, we need to check that the time-invariant characteristics of the participants in our sample did not change dramatically between the two waves to perform a valid DiD analysis. Two-sided T-tests and Chi-square tests are performed to compare the characteristics of the primary carers and recipients in our treatment and control group between the waves.

We perform multiple robustness tests for our carer outcomes, including sensitivity to the region-level fixed effects, cut-off dates and different sample restrictions. In order to minimise the impact of different characteristics of the sample on the results, in one of the robustness checks, we use propensity score weighting to weight the entire sample from Wave 15 and the control group from Wave 18 against the treatment group of Wave 18 for characteristics including age, sex of the carers, and the sex, disability status, disability type, marital status, education level, whether born in Australia, number of bedrooms, number of adults without disability in the household, and the rurality (Accessibility/Remoteness Index of Australia or ARIA) of the recipients of care’s residence. We bootstrapped the sample 1,000 times to account for the uncertainty around the weights and the estimates of the weighted sample. We also interacted individual characteristics with the treatment variable (𝑙𝑙𝑙𝑙_𝑙 #T_𝑙𝑙) and estimated the average effect to account for potential heterogenous treatment effects.

We also ran a placebo analysis with primary carers of people with disability who are not eligible for the NDIS because they are older than age 65 or who have mild disability using our treated and control group LGA. This estimates whether there is any effect for those not targeted by the NDIS.

Minimum sample size estimation

Given that the SDAC survey was the most comprehensive source of information available to understand different effects of the NDIS, we wanted to understand how big the SDAC sample frame needed to be in order to detect economically meaningful effect of the NDIS.

We therefore hypothesized several scenarios of different impact sizes and computed the minimum cluster size needed to detect these potential impacts.

SDAC’s sampling is clustered at the LGA level. We therefore computed cluster size conditional on the maximum number of LGA possible. There were 407 LGAs in Australia in 2018 - excluding the state of Western Australia which had its own NDIS trial at different periods. We also excluded three LGAs in other states because they were also the early trial sites (Hunter NSW, Barwon VIC and Barkly NT) and therefore 404 LGAs remained.

We assumed an equal number of LGAs in the control and treatment groups to compute the minimum number of people per LGA under different assumptions of intraclass correlations that would be needed to detect meaningful changes in outcomes. We used the means and standard deviations of the control group in the post treatment period, as a reference point for all computations - last column in Table 1. All results are for an 80% power test using a 5% two-sided test.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive Statistics (N = 1,054).

https://doi.org/10.1371/journal.pone.0321377.t001

Descriptive statistics

Table 1 presents the summary statistics of the sample by the NDIS participation group and wave, and Appendix S4 Table shows the results of our comparison tests. Of the 1,054 primary carers and recipient dyads, 439 are in the treatment group observed in wave 2015, 297 are observed in wave 2018, 182 are in the control group observed in wave 2015, and 136 are observed in wave 2018. The proportion of male recipients is higher in the control group observed in 2018 than in the other groups. There are differences in the number of bedrooms among the groups, although the effect size is small (max 0.3 rooms). People are more likely to live in major cities if they are in the treatment group. The average age of the primary carer was two years older for both the treatment and the control groups in 2015 than in 2018. There are fewer adults without disability in the household in the treatment group observed in 2015 than in other groups.

Main results

The estimated impact of the NDIS roll-out on the primary carers’ outcomes and the formal service used by people with disability are shown in Table 2. The coefficient for the variable NDIS available area# Wave18 reflects the absolute additional change in the outcomes between 2015 and 2018 for those who live in the LGAs where the NDIS had been rolled out between 2015 and 2018 (the treatment group) versus where the NDIS had not yet been rolled out (the control group). The outcomes include (1) overall number of formal services, (2) the extensive margin of formal services, (3) the intensive margin of the formal services, (4) weekly caring hours, (5) employment status, (6) social participation without recipients, (7) social participation with or without recipients. The complete list of covariates and results are provided in Appendix S5 Table.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. The association between NDIS availability in an LGA and changes in the outcomes of interest.

https://doi.org/10.1371/journal.pone.0321377.t002

We did not detect statistically significant NDIS impacts on any of the outcomes. The results show that the people with a profound or severe disability (or needing aids) who live in areas where the NDIS became available are, on average, nearly 10% less likely to use formal services compared to their counterparts in areas where the NDIS had not yet been implemented after the reform; however, the result is insignificant. For those who use formal services at least once per year, the NDIS reform did not lead to a statistically significant difference in the frequency of formal service use between the treatment and the control group.

For the carer outcomes, the weekly caring hours of primary carers increased by 4.8 hours between 2015 and 2018 for the carers in the NDIS area group relative to the control group, though it was not statistically significant. We find that primary carers in the treatment group are five percentage points less likely to have part-time or full-time work after the NDIS roll-out (β= -0.053, SE = 0.081) compared to the control group, but this result is not statistically significant. We find no statistically significant impact on participation in any social activity (β = -0.067, SE = 0.079) or participation in social activity without the recipient (β = -0.097, SE = 0.063). We then explore if the decrease in social participation is driven by the change in the social participation of recipients of care. This may reduce the chance of the primary carer accompanying the recipient to those events. The results (Appendix S6 Table) do not show a statistically significant effect (β = 0.101, SE = 0.115).

While previous reports have suggested possible gains for carers of young people, our subgroup analysis shows no significant heterogeneity in results for the 15-to-24-year-old age group (Appendix S7 Table).

Robustness checks and placebo test

We conduct multiple analyses to explore the sensitivity of our results on the carer outcomes to various levels of fixed effects, cut-off dates, sample restrictions and sample weightings. A summary of the robustness test results is presented in Table 3, and fully separated results are shown in the appendix. Generally speaking, all the results consistently show no significant effects. Although some of the results changed sign, as they stay statistically insignificant, it could be driven by random variation rather than meaningful effects because of the limitation of small sample size. First, due to the concern that the characteristics of our 2018 sample may differ from the characteristics of our 2015 sample, we weighted our sample based on the characteristics of the NDIS participants in the 2018 sample as described in Section 3.6 and bootstrapped the sample 1,000 times. The results are similar to our main analysis using the unweighted sample. We do not find statistically significant changes across the seven outcomes. (Appendix S8 Table). We then explore if the fixed effect at the regional level instead of the LGA level would impact the statistical power of our analysis. Since the NDIS roll-out is based on the regional level information, then broken down to the LGA level in our main analysis, the treatment and control groups remain identical. With the regional fixed effects, we find similar trends with a negative coefficient for employment and social participation and an increase in caring hours. All the results are statistically insignificant (Appendix S9 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Robustness checks and placebo test: estimated association between NDIS availability and outcomes under alternative assumptions.

https://doi.org/10.1371/journal.pone.0321377.t003

The primary carer may suffer short-term negative impact when the NDIS is being coordinated. Therefore, we also explore if the results differ for those whose areas that had NDIS available for at least one year. We vary the cut-off date for those considered to be treated from 30 June 2018–30 June 2017 when constructing T_𝑙𝑙, and keep the same control group to ensure that the differences in the results are not due to the change in the control groups. The results in Appendix S10 show that people living in areas where NDIS was available before June 2017 were 14% less likely to participate in social activities with or without the recipients of care, and the result is significant at the 10% level (β = -0.135, SE = 0.071) (Appendix S10 Table)

Primary carers may spend more time caring for the main recipients, and therefore, the effect size may become larger if the sample is restricted to main recipients only. We restrict our sample to the primary carer and main recipient relationship by excluding the non-main recipients in our sample (7.0%). The results are similar to our main analysis except that the possibility of participating in social activities with or without the recipients becomes significant at the 10% level (β = -0.115, SE = 0.065) (Appendix S11 Table).

About 10% of our primary carers in the sample are themselves eligible for NDIS and live in LGAs where the NDIS was available between 2015 and 2018. Therefore, there is a concern about whether the carer’s potential participation in the NDIS would also impact caring hours. We ran a sensitivity analysis restricting our sample to carers not eligible for the NDIS or living in LGAs where the NDIS is unavailable during the treatment period. Our main results were robust to this approach (Appendix S12 Table).

We also examine the robustness of the results with a different control group—primary carers of recipients who are unlikely to be eligible for the NDIS (because of age and disability severity level) and live in the same areas as the treatment group in our main analysis (areas with NDIS). This was done because we could not test for the parallel trend assumption, and we expected the results to be similar because the NDIS should also not affect this alternative control group. All the results remain statistically insignificant, although the outcomes of employment, social activities, and the frequency of using formal services are slightly above zero (Appendix Table S13).

Lastly, we examine the heterogeneous treatment effects by interacting the carer and recipient characteristics with our treatment term. We find no significant change to our main results (Appendix Table S14)

The results of the placebo analysis with the population of people with mild or moderate disability or older than 65 show no statistical changes in the outcomes. It indicates that the NDIS reform has no or minimal effects on the population, which is not targeted, consistent with our expectations (Appendix Table S15).

Minimum sample size estimation

We wanted to understand for different possible impacts, how big SDAC would have needed to be to detect meaning changes in the outcomes. The sample means in Appendix Figs S2a–S2e provide a graphical representation of the minimum sample size for three possible impacts, and considers a range of values of intraclass correlations for all outcomes. For example, taking S2a Fig, in blue we see that the minimum sample size needed to detect a difference of one hour of formal services per week. This effect size cannot be substantiated as significant when rho is larger than 0.1 as can be seen in the figure. For smaller rhos there may still need over 100 people per cluster to detect a statistically significant one-hour difference. Given that we have on average five observations per cluster, the SDAC survey would need to be much larger to detect a difference of one hour per week. S2c Fig shows that to detect a 10-percentage-point increase in employment we would only need a sample size of approximately 50 observations in each group even for large intraclass correlations. Generally speaking, the SDAC survey which is currently the only survey capable of evaluating the impact of the NDIS on carers would have benefited from having a substantial increase in sample size prior to the NDIS roll out sot that this multi-billion-dollar scheme could have been properly evaluated.

Limitation

Our research results are robust to different sample restrictions. However, there are several limitations. Even with some of the best available data on those with a disability, we suffer from the lack of statistical power in our analysis, which makes it susceptible to Type II error, or we are less likely to detect small changes in the outcomes accurately. Our sample size power calculation indicates that a much larger sample size would have been required to evaluate the NDIS using the same strategy. Future evaluations might usefully consider linked administrative data to be able to detect meaningful differences due to the NDIS implementations. However, such analyse will not be able to cover as broad a range of outcomes as the SDAC and ideally SDAC would have been increased prior to the NDIS implementation. Second, due to data limitations, we have only two time points and, therefore, cannot test the ‘parallel trend’ assumption in the pre-treatment period. However, our results are still robust when using alternative control groups. Third, we used a repeated cross-sectional dataset in our analysis, recruiting new respondents in each wave of the survey, which could impact the results if the composition of the samples changed. Not following up with the same population over time also means we cannot apply individual fixed effects in the model to control for time-invariant individual characteristics. However, we compared the characteristics of primary carers and their recipients of care in these two waves and controlled for the potential differences in these characteristics in our analysis. We have also conducted robustness checks with a matched sample and found similar results. We also explored the potential heterogeneity of the impact of NDIS, but we did not find any effect. Lastly, we do not have the NDIS utilisation data available, so we cannot account for the different impacts the utilisation rate may bring. But since the sample comprises people with disability on and not on the NDIS, we have applied the LGA fixed effects to control for areas characteristics that are stable over time, including factors on the supply side.

Further research

Future research should be conducted to evaluate the long-term effect of the NDIS when data becomes available. It is worth exploring further how we should support people with disability not on the NDIS in parallel with the NDIS. Access to disability services, thin markets and cultural awareness among service providers in Australia and also in international context are some topics that are worth exploring in future. Longitudinal data, and administrative data connecting with survey data capturing a wider range of outcomes should also be employed. The recent development of the National Disability Data Asset (NDDA) in Australia provides a unique opportunity to have a more complete picture of the programs and services people with disability use and thereby help better understand the overall impact of the NDIS as well as other policies aim at better supporting people with disability and their carers. Future research could also incorporate qualitative methods, such as interviews and focus groups, to gain deeper insights into the personal experiences and challenges faced by individuals accessing disability services, thereby complementing and enriching the quantitative findings and offering more insights into potential policy implications.